Finalmente, calcular el punto medio de cada segmento para visualizar la distancia como texto en la imagen\u200a\u2014\u200amidpoint y calculate_slope.<\/li>\n<\/ul>\n<\/li>\n<\/ul>\ndef calculate_coord(bbox, width, height):\n \"\"\"Return boxes coordinates\"\"\"\n xmin = bbox[1] * width\n ymin = bbox[0] * height\n xmax = bbox[3] * width\n ymax = bbox[2] * height\n\n return [xmin, ymin, xmax - xmin, ymax - ymin]<\/pre>\n\n\ndef calculate_centr(coord):\n \"\"\"Calculate centroid for each box\"\"\"\n return (coord[0]+(coord[2]\/2), coord[1]+(coord[3]\/2))\n\ndef calculate_centr_distances(centroid_1, centroid_2):\n \"\"\"Calculate the distance between 2 centroids\"\"\"\n return math.sqrt((centroid_2[0]-centroid_1[0])**2 + (centroid_2[1]-centroid_1[1])**2)\n\ndef calculate_perm(centroids):\n \"\"\"Return all combinations of centroids\"\"\"\n permutations = []\n for current_permutation in itertools.permutations(centroids, 2):\n if current_permutation[::-1] not in permutations:\n permutations.append(current_permutation)\n return permutations\n\n def midpoint(p1, p2):\n \"\"\"Midpoint between 2 points\"\"\"\n return ((p1[0] + p2[0])\/2, (p1[1] + p2[1])\/2)\n\n def calculate_slope(x1, y1, x2, y2):\n \"\"\"Calculate slope\"\"\"\n m = (y2-y1)\/(x2-x1)\n return m<\/code><\/pre>\n\n\nAhora que hemos definido estas funciones, podemos crear la funci\u00f3n principal llamada show_inference el cual ejecutar\u00e1 la predicci\u00f3n para las im\u00e1genes y graficar\u00e1 los recuadros y las distancias entre los peatones.<\/p>\n
La primera parte de esta funci\u00f3n es responsable de obtener la imagen y ejecutar la inferencia. Obtendremos un diccionario output_dict con las coordenadas de los objetos que el modelo ha detectado.<\/p>\n\n\n
image = Image.open(image_path)\n\n# the array based representation of the image will be used later\nimage_np = load_image_into_numpy_array(image)\n\n# Expanding dimensions\n# Since the model expects images to have shape: [1, None, None, 3]\nimage_np_expanded = np.expand_dims(image_np, axis=0)\n\n# Actual detection.\noutput_dict = run_inference_for_single_image(image_np, detection_graph)<\/code><\/pre>\n\n\nSeguidamente, definiremos un umbral confidence_cutoff=0.5 para evitar graficar predicciones con una probabilidad baja. En paralelo, obtendremos el tama\u00f1o de nuestra imagen y deberemos definir una relaci\u00f3n \u201cpixel-metro\u201d<\/strong> para calcular la distancia correctamente. He analizado algunas im\u00e1genes y he considerado que width\u200a\u2014\u200a150px = 7 meters es una buena relaci\u00f3n. Esta parte es complicada porque no tenemos en cuenta la perspectiva ni el \u00e1ngulo de la c\u00e1mara, es un problema complicado de generalizar para todas las im\u00e1genes y os animo a mejorarlo y mostrar vuestra soluci\u00f3n con nosotros ?.<\/p>\n\n\n# Get boxes only for person\nconfidence_cutoff = 0.5\nboxes, scores, classes = filter_boxes(confidence_cutoff, output_dict['detection_boxes'], output_dict['detection_scores'], output_dict['detection_classes'], [1])\n\n# Get width and heigth\nim = Image.fromarray(image_np)\nwidth, height = im.size\n\n# Pixel per meters - THIS IS A REFERENCE, YOU HAVE TO ADAPT THIS FOR EACH IMAGE\n# In this case, we are considering that (width - 150) approximately is 7 meters\naverage_px_meter = (width-150) \/ 7\n<\/code><\/pre>\n\n\nFinalmente, podemos calcular todos los centroides para nuestras predicciones y generar las combinaciones entre ellos. Posteriormente, podremos crear las l\u00edneas dx,dy que conectan los centroides y graficar la distancia.<\/p>\n\n\n
# Calculate normalized coordinates for boxes\ncentroids = []\ncoordinates = []\nfor box in boxes:\n coord = calculate_coord(box, width, height)\n centr = calculate_centr(coord)\n centroids.append(centr)\n coordinates.append(coord)\n\n# Calculate all permutations\npermutations = calculate_perm(centroids)\n\n# Display boxes and centroids\nfig, ax = plt.subplots(figsize = (20,12), dpi = 90)\nax.imshow(image, interpolation='nearest')\n\nfor coord, centr in zip(coordinates, centroids):\n ax.add_patch(patches.Rectangle((coord[0], coord[1]), coord[2], coord[3], linewidth=2, edgecolor='y', facecolor='none', zorder=10))\n ax.add_patch(patches.Circle((centr[0], centr[1]), 3, color='yellow', zorder=20))\n# Display lines between centroids\nfor perm in permutations:\n dist = calculate_centr_distances(perm[0], perm[1])\n dist_m = dist\/average_px_meter\n\n print(\"M meters: \", dist_m)\n middle = midpoint(perm[0], perm[1])\n print(\"Middle point\", middle)\n\n x1 = perm[0][0]\n x2 = perm[1][0]\n y1 = perm[0][1]\n y2 = perm[1][1]\n\n slope = calculate_slope(x1, y1, x2, y2)\n dy = math.sqrt(3**2\/(slope**2+1))\n dx = -slope*dy<\/code><\/pre>\n\n\nEn resumen, la funci\u00f3n deber\u00eda quedar tal que as\u00ed:<\/p>\n\n\n
import matplotlib.pyplot as plt\nimport matplotlib.patches as patches\nfrom random import randrange \n\n# If you want to test the code with your images, just add path to the images to the TEST_IMAGE_PATHS.\nPATH_TO_TEST_IMAGES_DIR = '\/gdrive\/My Drive\/Projects\/Pedestrian_Detection\/\nmodels\/research\/test_images'\nTEST_IMAGE_PATHS = [ os.path.join(PATH_TO_TEST_IMAGES_DIR,\n 'image{}.jpg'.format(i)) for i in range(8, 9) ] \n\n# Size, in inches, of the output images.\nIMAGE_SIZE = (12, 8)\n\n def show_inference(image_path):\n image = Image.open(image_path)\n # the array based representation of the image will be used later in order to prepare the\n # result image with boxes and labels on it.\n image_np = load_image_into_numpy_array(image)\n # Expand dimensions since the model expects images to have shape: [1, None, None, 3]\n image_np_expanded = np.expand_dims(image_np, axis=0)\n # Actual detection.\n output_dict = run_inference_for_single_image(image_np, detection_graph)\n\n # Get boxes only for person\n confidence_cutoff = 0.5\n boxes, scores, classes = filter_boxes(confidence_cutoff,\n output_dict['detection_boxes'], output_dict['detection_scores'],\n output_dict['detection_classes'], [1])\n\n # Get width and heigth\n im = Image.fromarray(image_np)\n width, height = im.size\n\n # Pixel per meters - THIS IS A REFERENCE, YOU HAVE TO ADAPT THIS FOR EACH IMAGE\n # In this case, we are considering that (width - 150) approximately is 7 meters\n average_px_meter = (width-150) \/ 7\n\n # Calculate normalized coordinates for boxes\n centroids = []\n coordinates = []\n for box in boxes:\n coord = calculate_coord(box, width, height)\n centr = calculate_centr(coord)\n centroids.append(centr)\n coordinates.append(coord)\n\n # Calculate all permutations\n permutations = calculate_perm(centroids)\n\n # Display boxes and centroids\n fig, ax = plt.subplots(figsize = (20,12), dpi = 90)\n ax.imshow(image, interpolation='nearest')\n for coord, centr in zip(coordinates, centroids):\n ax.add_patch(patches.Rectangle((coord[0], coord[1]), coord[2], coord[3], linewidth=2, edgecolor='y', facecolor='none', zorder=10))\n ax.add_patch(patches.Circle((centr[0], centr[1]), 3, color='yellow', zorder=20))\n\n # Display lines between centroids for perm in permutations:\n dist = calculate_centr_distances(perm[0], perm[1])\n dist_m = dist\/average_px_meter\n\n print(\"M meters: \", dist_m)\n middle = midpoint(perm[0], perm[1])\n print(\"Middle point\", middle)\n\n x1 = perm[0][0]\n x2 = perm[1][0]\n y1 = perm[0][1]\n y2 = perm[1][1]\n\n slope = calculate_slope(x1, y1, x2, y2)\n dy = math.sqrt(3**2\/(slope**2+1))\n dx = -slope*dy\n\n # Display randomly the position of our distance text\n if randrange(10) % 2== 0:\n Dx = middle[0] - dx*10\n Dy = middle[1] - dy*10\n else:\n Dx = middle[0] + dx*10\n Dy = middle[1] + dy*10\n\n if dist_m < 1.5:\n ax.annotate(\"{}m\".format(round(dist_m, 2)), xy=middle, color='white', xytext=(Dx, Dy), fontsize=10,\n arrowprops=dict(arrowstyle='->', lw=1.5, color='yellow'), bbox=dict(facecolor='red', edgecolor='white', boxstyle='round', pad=0.2), zorder=30)\n ax.plot((perm[0][0], perm[1][0]), (perm[0][1], perm[1][1]), linewidth=2, color='yellow', zorder=15) elif 1.5 < dist_m < 3.5:\n ax.annotate(\"{}m\".format(round(dist_m, 2)), xy=middle, color='black', xytext=(Dx, Dy), fontsize=8,\n arrowprops=dict(arrowstyle='->', lw=1.5, color='skyblue'), bbox=dict(facecolor='y', edgecolor='white', boxstyle='round', pad=0.2), zorder=30)\n ax.plot((perm[0][0], perm[1][0]), (perm[0][1], perm[1][1]), linewidth=2, color='skyblue', zorder=15)\n else:\n pass\n # Make prediction\n for file in TEST_IMAGE_PATHS:\n show_inference(file)<\/code><\/pre>\n\n\nDespu\u00e9s de todo el trabajo, podremos correr nuestro c\u00f3digo en im\u00e1genes. Simplemente debemos a\u00f1adir las im\u00e1genes a esta carpeta: ..\/Projects\/Pedestrian_Detection\/models\/research\/test_images<\/p>\n
_7coQNS7_14BXFUfCMphJF<\/p>\n
![\"\"](\"https:\/\/immune.institute\/wp-content\/uploads\/2020\/07\/1_7coQNS7_14BXFUfCMphJFw.jpeg\")
<\/p>\n
![\"\"](\"https:\/\/immune.institute\/wp-content\/uploads\/2020\/07\/1_p7zZ6TwaPH9omVDW0ustww.jpeg\")
<\/p>\n
Wow!<\/strong> Tienen Buena pinta y podr\u00eda incluso mejor si ejecutamos este c\u00f3digo en videos. As\u00ed que, \u00a1vamos all\u00e1!<\/p>\nPredicci\u00f3n en v\u00eddeos<\/strong><\/h3>\nEl c\u00f3digo para ejecutar el modelo en v\u00eddeos es igual que para las im\u00e1genes porque usaremos OpenCV para dividir el video en frames y as\u00ed poder procesar cada frame como im\u00e1genes individuales.<\/p>\n\n\n
import cv2\nimport matplotlib\nfrom matplotlib import pyplot as plt\nplt.ioff()matplotlib.use('Agg')\n\nFILE_OUTPUT = '\/gdrive\/My Drive\/Projects\/Pedestrian_Detection\/models\/research\/test_\nimages\/Rail_Station_Predicted2.mp4'\n\n# Playing video from file\ncap = cv2.VideoCapture('\/gdrive\/My Drive\/Projects\/Pedestrian_Detection\/\nmodels\/research\/test_images\/Rail_Station_Converted.mp4')\n\n# Default resolutions of the frame are obtained.The default resolutions are system dependent.\n# We convert the resolutions from float to integer.\nwidth = int(cap.get(3))\nheight = int(cap.get(4))\n\ndim = (width, height)\nprint(dim)\n\ntope = 10\ni = 0\nnew = True\nwith detection_graph.as_default():\n with tf.Session(graph=detection_graph) as sess:\n # Definite input and output Tensors for detection_graph\n image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')\n # Each box represents a part of the image where a particular object was detected.\n detection_boxes = detection_graph.get_tensor_by_name('detection_boxes:0')\n\n # Each score represent how level of confidence for each of the objects.\n # Score is shown on the result image, together with the class label.\n detection_scores =\n detection_graph.get_tensor_by_name('detection_scores:0')\n detection_classes =\n detection_graph.get_tensor_by_name('detection_classes:0')\n num_detections =\n detection_graph.get_tensor_by_name('num_detections:0')\n\n i = 0\n while(cap.isOpened()):\n # Capture frame-by-frame\n ret, frame = cap.read()\n if ret == True:\n # Correct color\n frame = gray = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)\n\n # Expand dimensions since the model expects images to have shape: [1, None, None, 3]\n image_np_expanded = np.expand_dims(frame, axis=0)\n # Actual detection.\n (boxes, scores, classes, num) = sess.run(\n [detection_boxes, detection_scores, detection_classes, num_detections],\n feed_dict={image_tensor: image_np_expanded})\n\n # Filter boxes\n confidence_cutoff = 0.5\n boxes, scores, classes = filter_boxes(confidence_cutoff,\n np.squeeze(boxes), np.squeeze(scores), np.squeeze(classes), [1])\n\n # Calculate normalized coordinates for boxes\n centroids = []\n coordinates = []\n for box in boxes:\n coord = calculate_coord(box, width, height)\n centr = calculate_centr(coord)\n centroids.append(centr)\n coordinates.append(coord)\n\n # Pixel per meters\n average_px_meter = (width-150) \/ 7\n \n permutations = calculate_perm(centroids)\n\n # Display boxes and centroids\n fig, ax = plt.subplots(figsize = (20,12), dpi = 90, frameon=False)\n ax = fig.add_axes([0, 0, 1, 1])\n ax.axis('off')\n ax.spines['top'].set_visible(False)\n ax.spines['right'].set_visible(False)\n ax.spines['bottom'].set_visible(False)\n ax.spines['left'].set_visible(False)\n ax.get_xaxis().set_ticks([])\n ax.get_yaxis().set_ticks([])\n for coord, centr in zip(coordinates, centroids):\n ax.add_patch(patches.Rectangle((coord[0], coord[1]),\n coord[2], coord[3], linewidth=2, edgecolor='y', facecolor='none', zorder=10))\n ax.add_patch(patches.Circle((centr[0], centr[1]), 3,\n color='yellow', zorder=20))\n\n # Display lines between centroids\n for perm in permutations:\n dist = calculate_centr_distances(perm[0], perm[1])\n dist_m = dist\/average_px_meter\n\n x1 = perm[0][0]\n y1 = perm[0][1]\n x2 = perm[1][0]\n y2 = perm[1][1]\n \n # Calculate middle point\n middle = midpoint(perm[0], perm[1])\n\n # Calculate slope\n slope = calculate_slope(x1, y1, x2, y2)\n dy = math.sqrt(3**2\/(slope**2+1))\n dx = -slope*dy\n\n # Set random location\n if randrange(10) % 2== 0:\n Dx = middle[0] - dx*10\n Dy = middle[1] - dy*10\n else:\n Dx = middle[0] + dx*10\n Dy = middle[1] + dy*10\n\n if dist_m < 1.5:\n ax.annotate(\"{}m\".format(round(dist_m, 2)), xy=middle,\n color='white', xytext=(Dx, Dy), fontsize=10, arrowprops=dict(arrowstyle='->',\n lw=1.5, color='yellow'), bbox=dict(facecolor='red', edgecolor='white',\n boxstyle='round', pad=0.2), zorder=35)\n ax.plot((perm[0][0], perm[1][0]), (perm[0][1], perm[1][1]),\n linewidth=2, color='yellow', zorder=15)\n elif 1.5 < dist_m < 3.5:\n ax.annotate(\"{}m\".format(round(dist_m, 2)), xy=middle,\n color='black', xytext=(Dx, Dy), fontsize=8, arrowprops=dict(arrowstyle='->',\n lw=1.5, color='skyblue'), bbox=dict(facecolor='y', edgecolor='white',\n boxstyle='round', pad=0.2), zorder=35)\n ax.plot((perm[0][0], perm[1][0]), (perm[0][1], perm[1][1]),\n linewidth=2, color='skyblue', zorder=15)\n else:\n pass\n\n ax.imshow(frame, interpolation='nearest')\n \n # This allows you to save each frame in a folder\n #fig.savefig(\"\/gdrive\/My\n Drive\/Projects\/Pedestrian_Detection\/models\/research\/test_images\/TEST_{}.png\".\nformat(i))\n # i += 1\n # Convert figure to numpy\n fig.canvas.draw()\n\n img = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8,\n sep='')\n img = img.reshape(fig.canvas.get_width_height()[::-1] + (3,))\n\n img = np.array(fig.canvas.get_renderer()._renderer)\n img = cv2.cvtColor(img,cv2.COLOR_RGB2BGR)\n\n if new:\n print(\"Define out\")\n out = cv2.VideoWriter(FILE_OUTPUT,\n cv2.VideoWriter_fourcc(*'MP4V'), 20.0, (img.shape[1], img.shape[0]))\n new = False\n\n out.write(img)\n else:\n break\n\n # When everything done, release the video capture and video write objects\n cap.release()\n out.release()\n\n # Closes all the frames\n cv2.destroyAllWindows()\n<\/code><\/pre>\n\n\nComo hicimos antes, simplemente debemos copiar nuestro video en: ..\/Projects\/Pedestrian_Detection\/models\/research\/test_images y actualizar el path en cap = cv2.VideoCapture(\u2026).Tambi\u00e9n podemos definir un nombre para FILE_OUTPUT.<\/p>\n\n\n\n","protected":false},"excerpt":{"rendered":"
Usando TensorFlow Object Detection API para detectar peatones y medir la distancia social entre ellos. Hoy, por desgracia, todo el mundo est\u00e1 familiarizado con el t\u00e9rmino \u00abdistancia social\u00bb. Es algo con lo que tendremos que vivir por un tiempo hasta que todo vuelva a la normalidad. En Immune Technology Institute hemos desarrollado una aplicaci\u00f3n usando TensorFlow Object Detection […]<\/p>\n","protected":false},"author":3,"featured_media":7981,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_crdt_document":"","footnotes":""},"categories":[1],"tags":[],"class_list":["post-8923","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blog"],"acf":[],"_links":{"self":[{"href":"https:\/\/immune.institute\/en\/wp-json\/wp\/v2\/posts\/8923","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/immune.institute\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/immune.institute\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/immune.institute\/en\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/immune.institute\/en\/wp-json\/wp\/v2\/comments?post=8923"}],"version-history":[{"count":0,"href":"https:\/\/immune.institute\/en\/wp-json\/wp\/v2\/posts\/8923\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/immune.institute\/en\/wp-json\/wp\/v2\/media\/7981"}],"wp:attachment":[{"href":"https:\/\/immune.institute\/en\/wp-json\/wp\/v2\/media?parent=8923"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/immune.institute\/en\/wp-json\/wp\/v2\/categories?post=8923"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/immune.institute\/en\/wp-json\/wp\/v2\/tags?post=8923"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
![\"\"](\"https:\/\/immune.institute\/wp-content\/uploads\/2020\/07\/1_4TbSdKWejM2r4WSlZNfI8g.jpeg\")
<\/p>\n![\"\"](\"https:\/\/principal.immune.institute\/wp-content\/uploads\/2020\/07\/0_X6Lm9b8_SRsCzFkV-213x300-1.jpeg\")
<\/p>\n![\"\"](\"https:\/\/immune.institute\/wp-content\/uploads\/2020\/07\/0_0YsPZGrOb14vQAeP.png\")
<\/p>\n![\"\"](\"https:\/\/immune.institute\/wp-content\/uploads\/2020\/07\/1_nV749oNHAeKWowUrSBsO3w.jpeg\")
<\/p>\n![\"\"](\"https:\/\/principal.immune.institute\/wp-content\/uploads\/2020\/07\/1_ofphZZWhUjTYOAoQG88DTA.jpeg\")
<\/p>\n